The Latency of a Domain-General Visual Surprise Signal is Attribute Dependent

Abstract

Predictions concerning upcoming visual input play a key role in resolving percepts. Sometimes input is surprising, under which circumstances the brain must calibrate erroneous predictions so that perception is veridical. Despite the extensive literature investigating the nature of prediction error signalling, it is still unclear how this process interacts with the functionally segregated nature of the visual cortex, particularly within the temporal domain. Here, we recorded electroencephalography (EEG) from humans whilst they viewed static image trajectories containing a bound object that sequentially changed along different visual attribute dimensions (shape and colour). Crucially, the context of this change was designed to appear random (and unsurprising) or violate the established trajectory (and cause a surprise). Event-related potential analysis found no effects of surprise after controlling for cortical adaptation. However, multivariate pattern analyses found whole-brain neural representations of visual surprise that overlapped between attributes, albeit at distinct, attribute-specific latencies. These findings suggest that visual surprise results in whole-brain, generalised (i.e., attribute-agnostic) prediction error responses that conform to an attribute-dependent temporal hierarchy.